Re: [67ATTENDEES] Anyone get the 'broadband' connection in the IETF hotel

Fred Baker writes:
> So you may find this interesting. Last night, at 6:30, I started a
> ping study from my room to a computer back at "the ranch". You may
> look at the pictures at
>       ftp://ftpeng.cisco.com/fred/sheraton/ping-study-11-7-rtt.htm

Cute, thanks!

My laptop is permanently monitored using Tobi Oetiker's SmokePing
tool, which has another nice way of visualizing delay distributions
(loss is also displayed, by the color of the median line):

http://lg.net.switch.ch/cgi-bin/smokeping.cgi?target=World.Europe.Switzerland.Simon.Laptop

The measurement server is in Switzerland, so the base RTT is >160ms
(and timescale is based on Central European Time).

> [...] The second is displayed on a log-linear chart, because the
> upper bound value is a 35 second RTT on a ping that happened a bit
> after 11:00 last night

Sudden load increase as busloads of IETFers returning from the
(excellent) social were unloaded onto the poor hotel network?

> and it would be really nice to be able to see the normal samples,
> which are in the 60-500ms range. The delay distribution chart is
> interesting in this way: the 90% confidence delay is shown in the
> chart; the last 10% has samples, as I said, all the way from 750 ms
> to 35 seconds. But instead of the typical Poisson distribution in
> which 90% of cases are within zero and twice the mean, there is a
> big bump around the mean and a really long and pronounced tail in
> which there is a very high probability that delay on a packet is
> many times the mean RTT. I would guess that your average TCP is
> probably unnecessarily retransmitting a significant percentage of
> its transmissions simply because of wildly varying delay.

Right.  This would argue for tuning the network differently, namely
for shorter queues, possibly (but not necessarily) at the expense of
higher loss.  Sounds like an excellent place for RED (remember RED?),
but even with tail-drop and just smaller buffer limits in some router,
it would probably work better for most if not all users.  If someone
has contacts with the operator of the congested part of the backbone
(AT&T?), I wouldn't mind talking to them...

> Basically, last night, there was a significant delay bump from 10:00
> PM to 1:30 AM and another one from about 6:00 AM to 8:00 AM.
> Surprise. 10:00 is about when we got back from the social, dinner,
> or whatever else we did last night, and 6:00-8:00 is when we are all
> in our rooms and haven't yet materialized downstairs for coffee.

> Something about having 1000 internet geeks walk into a hotel that is  
> designed for executive travelers and wonder what's happening back  
> home :-)

I can easily imagine that this type of behavior "trickles down" to the
non-geek traveler population.
-- 
Simon.

_______________________________________________
67ATTENDEES mailing list
67ATTENDEES@ietf.org
https://www1.ietf.org/mailman/listinfo/67attendees